From Pixels to Prediction: Reviewing the Role of Artificial Intelligence in Body Composition Analysis

Evaluation of body composition (BC) is a set of biomarkers, including fat, muscle and bone, that allows the quantification of an individual's composition at different levels of complexity (whole body, tissues, macroscopic and microscopic). Recently, BC has gained medical interest due to its impact on health outcomes in both oncologic and non-oncologic conditions. BC parameters are also valuable for opportunistic screening, offering benefits to patients in prevention and health management. Among non-invasive methods to assess BC, cross-sectional imaging, for example, tomography (CT) and magnetic resonance (MR), can provide quantitative data from exams conducted for other reasons that can significantly aid large-scale health prevention. Recent studies highlight BC's potential in personalized medicine, but challenges like lengthy and poorly repeatable manual segmentation and complex analysis have limited its routine clinical use. Artificial intelligence (AI) can address these issues by simplifying and automating the process, from segmentation to prediction (e.g., automatic muscle segmentation for sarcopenia assessment and AI assessment of BC as opportunistic evaluation during CT scan for other clinical reasons). This review aims to summarize recent advancements in BC and AI, showcasing their synergy in enhancing the management of health conditions, from diagnosis to personalized treatments, while discussing current limitations and future challenges.